What problem does it solve? In a shared-tree repository where multiple AI agents and a human collaborate, it is easy to make changes without knowing who must approve them or which gate enforces the rule. This Skill answers the question "I am about to change X — what procedure applies, who signs off, and where does the record go?" before any change is made. ## Core Features & Use Cases - Change-class routing table: Maps 13 change classes (design intent, UI surfaces, fiction text, balance numbers, maps, PRD edits, save schemas, commits, pushes, releases, deferred work, milestone claims, human TODO lists) to their sign-off authority, enforcement gate, and record location. - Standing approvals vs always-human calls: Distinguishes actions agents may perform autonomously (checkpoint pushes, schema bumps with migrations) from human-only decisions (releases via /ship, taste picks, force-pushes, save wipes). - ADR authoring mechanics: Covers reserving ADR numbers via tree-claim, writing entries in the live band, superseding old decisions without deleting history, and the precedence rules between living docs and ADRs. - Use Case: Before editing a balance slider export, an agent consults this Skill, learns that feel numbers are human-only via the cockpit slider, and instead transcribes the human's exported diff while running pnpm run verify:balance. ## Quick Start Ask the agent to load the kami-change-control skill and tell you what procedure and sign-off apply before committing your planned change to the shared tree.
